At lunch, Nick Schulz asked me what Robert Hall is known for. I said that Hall changed my view of macroeconomics. Even today, Hall’s work influences how I think about global warming. So pull up a chair, grab a cup of your favorite beverage, and let me reminisce.Greg Mankiw has an essay called The Scientist and the Engineer, which is about the gulf between academic macro and the macro used by forecasters and policymakers. I experienced this conflict intensely early in my career.
From January through August of 1976, I was a research assistant in the National Income section at the Federal Reserve Board. I worked for Dave Wyss, a model jockey. A model jockey might ask, “What will be the effect of the stimulus proposal on GDP?” You have an equation, estimated using past data, that predicts how much consumption will go up for a given increase in disposable income. You interact that with a bunch of other equations, and out comes your answer. That is how the engineers look at macro.
In the fall of 1976, I started grad school at MIT. I was surprised to find that young scientists were not impressed by engineers. Ray Fair, an engineer-type from Yale, came to teach a course in macro-econometric modeling, and he became a laughing-stock. The course consisted of Fair walking through his own model of about 100 equations, and I can remember classmate John Huizinga (now at U. of Chicago) sarcastically commenting, “Let me guess. The lagged dependent variable, right?” In layman’s terms, what Huizinga was complaining about was that regardless of the type of variable that Fair was trying to predict (inflation, unemployment, consumer spending, what have you), he always found a theoretical rationale for including the previous quarter’s value of that variable in the prediction equation. The statistical performance of his forecasts rested on this crutch.
Nonetheless, I was still feeling loyalty to the engineers. In fact, a lot of what the scientists were doing struck me as equivalent to asking how many angels can dance on the head of a pin. I remember lashing out at Stan Fischer at the beginning of one his third lecture on “monetary growth models,” which felt to me like angel-counting. I remember that incident primarily because I feel very sorry about it. Fischer, now head of Israel’s central bank, is one of the real gentlemen of the profession.
But at one point, Bob Hall gave a seminar, with the awkward title of “The Life Cycle/Permanent Income Hypothesis of Consumption.” The gist of it was that the quarterly change of income had no statistically significant relationship to the quarterly change of consumption. You might see data on income and consumption that look like this:
Year Q inc con
1968 1 100 80
1968 2 105 83
1968 3 109 88
1968 4 115 92
and think that consumption depends positively on income. The bigger the income number, the bigger the consumption number.
But if you look at the quarterly changes, you get
Year Q inc con
1968 2 $5 $3
1968 3 $4 $5
1968 4 $6 $4
Looked at this way, one does not find that the bigger the income number, the bigger the consumption number.
Lots of people, myself included, suspected Hall of pulling some sort of Swindle. It took him a few years to get his paper published. Soon, I got my degree and went back to work at the Fed.
When I rejoined the engineers at the Fed, Hall had planted seeds of doubt in my mind about the reliability of looking at data over time in terms of absolute levels, because looking at quarterly changes can produce different results. Clive Granger (who recently won a Nobel) and Christopher Sims were raising all sorts of alarms about the problem of what is called nonstationary data. After a few years of agonizing about this issue off and on, I came to the conclusion that the engineers’ models are an exercise in self-deception. Macroeconomics is steeped in data but issues in macro cannot be resolved by data. One’s views on macro are more like religion.
I have not kept up with what the scientists of academia are doing in macro. It still strikes me as angel-counting. But I don’t believe that the engineers know very much, either. I don’t think of the economy in terms of a system of equations. I think that what Ben Bernanke is doing now–floundering, trying to prop up the financial system without creating too much inflation or moral hazard–is more typical of how one has to operate as a central banker (unless things are going smoothly, in which case you sit back and try to do as little as possible).
Hall’s methods also affect my outlook on global warming. The fact that global temperatures are high at the same time that atmospheric carbon dioxide is high does not convince me. Changes in the two over shorter time horizons appear to be unrelated. I understand that there are scientific arguments that would justify a strong long-term relationship and a weak short-term relationship, but those arguments only serve to convince me that the observed data cannot rule out man-made global warming. The case that the data prove man-made global warming is very weak, the way that I see it.
I don’t know what Bob Hall himself thinks about global warming. Whatever his views are, I would take his opinions seriously. He has an excellent mind.
READER COMMENTS
E. Barandiaran
Sep 13 2008 at 5:40am
Apparently some macro engineers have opted to have fun with statistics (see the two references cited by G. Mankiw in his post
http://gregmankiw.blogspot.com/2008/09/fun-with-statistics.html).
Unfortunately Tyler Cowen and others have lost their sense of humor: they condemn Donald Luskin who laughs at those that are having fun with statistics.
Richard
Sep 13 2008 at 7:43am
Arnold’s observation on climate change suggests that Sarah Palin may have been influenced by Robert Hall’s view as well.
Devin Finbarr
Sep 13 2008 at 10:58am
Arnold,
I know you’re a Mencius Moldbug fan. Did you read his latest article on how modern economists should be considered astrologists? I found it very convincing. What was your take?
Mike Sproul
Sep 13 2008 at 11:24am
Why is macroeconomic theory so bad? Because 200 years ago, David Ricardo and Henry Thornton convinced people that paper money can have value even if it has no backing. Ever since, when tight money conditions have caused recessions, their followers have argued that there is no point in increasing the money supply by x%, because that will only increase the price of groceries by x%. The necessary new money was not issued and the recessions continued, with mainstream economists insisting that the market will correct itself if left alone. Small wonder that dissident views were spawned, including those built on silly tautologies like Y=C+I+G, or MV=Py.
The first step macroeconomists must take is to recognize that the dollar is not fiat money. The value of the dollar is equal to the value of the assets backing those dollars. If more dollars are needed to ease tight money conditions, those dollars can safely be issued as long as the Fed’s assets increase in step with the number of dollars.
dearieme
Sep 13 2008 at 1:48pm
“there are scientific arguments that would justify a strong long-term relationship”: not really. It’s a logarithmic dependence so once CO2 concentrations are as high as they now are, increasing them further would, by itslef, have a negligible influence. All the climate modelling of which you read is an exercise in fiddling with purported amplifiers to introduce some positive feedback. And the measured temperatures seem to be potentially so biased, and actually so fudged, that they cast little light on the business.
Steve Roth
Sep 13 2008 at 3:02pm
Post hoc ergo propter hoc is the eternal comeback in both scientific and mechanical macro, because we can’t replay the world a different way and see what “would have” happened.
It’s a very effective counter when you’re looking at brief periods (let’s call it micromacro) with no comparative or control element.
But:
1. That comeback is much less convincing when comparing outcomes from two sets of policies
A. Over a long period
and
B. When those policies are consistently and systematically different
2. Arnold is exactly right that (long-term) correlational data can quite effectively disprove a theory. (Likewise, Mankiw stipulates to this in his “Growth of Nations.”)
i.e.: Growth in Europe and the US have been the same over decades, while tax rates have been massively different (40% versus 28% of GDP).
i.e. 2: Since WWII, growth under Democratic presidents has been far greater.
These facts quite effectively disprove the theory that Republican/supply-side/trickle-down policies cause greater/faster economic growth.
Aside: on the two “equality” studies that Mankiw post-hoc pooh-poohs.
Dems deliver less rich/poor inequality: Blinder’s point is that everyone is better off under Democratic policies. The equality is a wonderful (causally connected?) bonus.
Pubs deliver less male/female inequality: women appear to have done relatively better under ‘pubs because working-class men have done so poorly under same.
Ben Kalafut
Sep 15 2008 at 5:13pm
You’re really almost reveling here in your ignorance of how physical science is done. The data never “prove” anything in the sense you’re looking for; they are merely compatible with or incompatible with this description or that.
Coinciding high CO2 levels and temperatures shouldn’t convince you, and that you would even feel the need to say that they don’t shows a contempt for physical science and physical scientists. Get it though your head: nowhere in the scientific literature is the coincidence in itself treated as evidence.
Climate models are based on physics that we know to be very good in predicting myriad phenomena, combined (of course) in somewhat of a novel way to attack this problem. These models are the missing bits, in addition to the high CO2 levels and measured temperatures and rates of warming (rates are important), that should have you convinced. That some economists have in the past made lousy models does not in any way mean that physicists and atmospheric scientists cannot make good ones, and it is, again, contemptuous of you to suggest otherwise. Just what flaw are you seeing in GCMs that the experts are not? Or will you simply admit that you are being irrational and superstitious.
Doesn’t Bryan Caplan say that the ignorant should simply admit that they are not entitled to an opinion? Maybe that advice should be taken.
dirtyrottenvarmint
Sep 16 2008 at 6:39pm
Roth:
If Europe and US GDP growth has been equivalent while tax rates differ, and if US GDP growth during Democrat presidential terms has been greater, this neither proves nor disproves anything. In order to “disprove” supply-side theory one would have to hold all other factors equal. This is why econometrics is BS.
One could just as easily deduce from the information you have provided that i) American productivity is less dependent on tax regimes than that of Europeans, and ii) the party affiliation of the President has little to do with economic growth (which would not be surprising given that little, if anything, in the Constitution gives the President any ability to affect economic fundamentals.)
dirtyrottenvarmint
Sep 16 2008 at 7:11pm
Kalafut:
Baloney, blarney, and balderdash. “Climate models are based on physics that we know to be very good in predicting myriad phenomena, combined (of course) in somewhat of a novel way to attack this problem,” you say. This statement rests on a definition of “very good” so liberal as to be virtually devoid of meaning. Climate models to date have excluded variables of potentially astronomical import, such as the effect of solar variation on cosmic ray-influenced low-level cloud formation. None of the models that have been fitted to past results have been at all predictive, and all of the models that predict future warming do so based not on a CO2-warming link but instead on a “positive feedback effect” for which there is absolutely no scientific basis whatsoever.
Economists creating multivariable models of economic growth and physicists creating multivariable models of climate change have at least one thing in common: neither is engaged in anything remotely approaching scientific research. The scientific method is a systematized model for inductive reasoning that rests on 3 legs: i) a testable hypothesis, ii) testing of that hypothesis through rigorous and controlled data collection, and iii) replication of test results by other scientists. A model can not be “peer tested” because regardless of who pushes the Enter key, the model is the same (does not satisfy (iii)) and furthermore a model is not a “test” of a real-world hypothesis because model results are not “data” (does not satisfy (ii)).
The relationship of CO2 and sunlight can be tested in controlled conditions; there is well-accepted evidence that CO2 absorbs some wavelengths of electromagnetic radiation but not others. From this it can be inferred that CO2 can potentially have some effect on climate. What that effect is has never been tested. There is no further scientific (hypothesis-tested-by-real-world-data-collected-under-controlled-conditions) evidence from which any additional information can be determined regarding the relationship between CO2 and climate, including the sign (+/-).
Comments are closed.